[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

WO2023029514A1 - 科室分诊方法、系统、设备以及存储介质 - Google Patents

科室分诊方法、系统、设备以及存储介质 Download PDF

Info

Publication number
WO2023029514A1
WO2023029514A1 PCT/CN2022/087997 CN2022087997W WO2023029514A1 WO 2023029514 A1 WO2023029514 A1 WO 2023029514A1 CN 2022087997 W CN2022087997 W CN 2022087997W WO 2023029514 A1 WO2023029514 A1 WO 2023029514A1
Authority
WO
WIPO (PCT)
Prior art keywords
department
model
prediction
appeal
effective
Prior art date
Application number
PCT/CN2022/087997
Other languages
English (en)
French (fr)
Inventor
周尚思
赵璐偲
Original Assignee
康键信息技术（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 康键信息技术（深圳）有限公司 filed Critical 康键信息技术（深圳）有限公司
Publication of WO2023029514A1 publication Critical patent/WO2023029514A1/zh

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/20—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the management or administration of healthcare resources or facilities, e.g. managing hospital staff or surgery rooms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/903—Querying
    • G06F16/9035—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/237—Lexical tools
    • G06F40/247—Thesauruses; Synonyms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
    • G06F40/295—Named entity recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Definitions

  • the embodiments of the present application relate to the field of smart medical technology, and in particular to a department triage method, system, device and storage medium.
  • Triage is a field of research in medical artificial intelligence and Internet medical care, and it is also the entry path for patients to see a doctor. Mistakes in triage will lead to a chain reaction of errors in the follow-up consultation process.
  • Intelligent triage usually means that hospitals or Internet medical institutions collect patients' descriptions (such as gender, age, and description information about their own physical abnormalities, etc.) with the consent of patients. ), and then the medical practitioners conduct a professional analysis of the chief complaint, and label the chief complaint to the corresponding department. After collecting a certain amount of labeled data, they use the deep learning algorithm to train the prediction model, and finally based on the model after the training is completed.
  • the embodiment of the present application provides a department triage method, system, device, and storage medium.
  • the coverage of department prediction results is higher, and the accuracy of department recommendation is also higher.
  • the embodiment of the present application provides a department triage method, the department triage method includes:
  • the effective appeal features are input into a plurality of different prediction models to perform department prediction respectively, and the department prediction results output by each of the prediction models are obtained;
  • the prediction models include a deep learning model, a rule matching model, a drug matching model, and disease matching model;
  • the information of the target department is pushed according to the sorting result.
  • the embodiment of the present application provides a department triage system, and the department triage system includes:
  • An information acquisition unit configured to acquire patient appeal information
  • the first processing unit is configured to extract effective appeals from the appeal information through a deep learning entity recognition model, and perform feature extraction on the effective appeals to obtain effective appeal features;
  • the department prediction unit is used to input the effective appeal features into a plurality of different prediction models to perform department prediction respectively, and obtain the department prediction results output by each of the prediction models;
  • the prediction models include deep learning models, rule matching model, drug matching model, and disease matching model;
  • the second processing unit is configured to filter and sort all the department prediction results to obtain the sorting results
  • the department recommending unit is configured to push target department information according to the ranking result.
  • an embodiment of the present application provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and operable on the processor, and when the processor executes the computer program, the following is realized: as described above Departmental triage method.
  • the embodiment of the present application provides a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are used to execute: the above-mentioned department triage method.
  • the first aspect of the embodiment of the present application provides a departmental triage method.
  • This method first obtains the patient's appeal information; and then judges the patient's appeal information to effectively appeal, and can filter out unclear or invalid information caused by mistouching , so as to ensure the accuracy of the follow-up prediction results; then extract the effective appeal features, and based on the effective appeal features, use the deep learning model, the rule matching model, the drug matching model and the disease matching model to make predictions respectively, and get each Compared with the prediction results of related schemes, the department prediction results predicted by the prediction model are single-label prediction results.
  • This method uses the deep learning model, rule matching model, drug matching model and disease matching model to make predictions respectively, and can obtain multi-label
  • the prediction results not only improve the coverage of the predicted departments, but also improve the accuracy of the predicted departments; finally, this method also filters and sorts the predicted results of all the predicted departments, which can further improve the accuracy of the predicted departments.
  • FIG. 1 is a schematic diagram of a system architecture for implementing a department triage method provided by an embodiment of the present application
  • Fig. 2 is a logic block diagram of a kind of department triage method provided by one embodiment of the present application
  • Fig. 3 is a schematic flowchart of a department triage method provided by an embodiment of the present application.
  • references to “one embodiment” or “some embodiments” described in the description of the embodiments of the present application mean that specific features described in conjunction with the embodiments of the present application are included in one or more embodiments of the embodiments of the present application. , structure or characteristics.
  • appearances of the phrases “in one embodiment,” “in some embodiments,” “in other embodiments,” “in other embodiments,” etc. in various places in this specification are not necessarily All refer to the same embodiment, but mean “one or more but not all embodiments” unless specifically stated otherwise.
  • the terms “including”, “comprising”, “having” and variations thereof mean “including but not limited to”, unless specifically stated otherwise.
  • the embodiment of the present application can acquire and process related data based on medical artificial intelligence technology.
  • artificial intelligence is the theory, method, technology and application of using digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.
  • Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation/interaction systems, and mechatronics.
  • Artificial intelligence software technology mainly includes computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning/deep learning.
  • the embodiments of the present application mainly relate to natural language processing technology and machine learning/deep learning technology in medical artificial intelligence.
  • Triage is a field of research in medical artificial intelligence and Internet medical care, and it is also the entry path for patients to see a doctor. Mistakes in triage will lead to a chain reaction of errors in the follow-up consultation process.
  • Intelligent triage usually means that hospitals or Internet medical institutions collect patients' descriptions (such as gender, age, and description information about their own physical abnormalities, etc.) with the consent of patients. ), and then the medical practitioners conduct a professional analysis of the chief complaint, and label the chief complaint to the corresponding department. After collecting a certain amount of labeled data, they use the deep learning algorithm to train the prediction model, and finally based on the model after the training is completed.
  • the prediction results of this scheme are mostly single-label results, not only the coverage of the predicted department is low, but also the accuracy of the predicted department is not high. Moreover, there are a large number of cases of unclear expressions or mistouched input in the chief complaints provided by patients. In related solutions, the unclear or false touches in the chief complaints will be ignored, thereby reducing the accuracy of subsequent prediction results.
  • the embodiment of the present application first obtains the patient's appeal information; and then judges the effective appeal of the patient's appeal information, which can filter out unclear or invalid information caused by mistaken touch, so as to ensure the accuracy of subsequent prediction results degree; then extract the effective appeal features, and based on the effective appeal features, use multiple different prediction models to predict the departments respectively, and obtain the prediction results of multiple departments.
  • the embodiment of this application can obtain multi-label prediction results, and can realize multi-label refined prediction, which not only improves the coverage of the prediction department, but also greatly improves the accuracy of the prediction department; Filtering and sorting the prediction results of each department can further improve the accuracy of the prediction department.
  • the department triage method provided by an embodiment of the present application can be implemented in an electronic device.
  • the terminal/device may be a mobile electronic device or a non-mobile electronic device.
  • Mobile electronic devices can be mobile phones, tablet computers, notebook computers, handheld computers, vehicle electronic devices, wearable devices, super mobile personal computers, netbooks, personal digital assistants, etc.; non-mobile electronic devices can be personal computers, televisions, teller machines or Self-service machines, etc.; the implementation plan of this application does not make specific limitations.
  • the electronic device may include a processor, an external memory interface, an internal memory, a universal serial bus (universal serial bus, USB) interface, a charging management module, a power management module, a battery, an antenna, a mobile communication module, a wireless communication module, an audio module, Speakers, receivers, microphones, headphone jacks, sensor modules, buttons, motors, indicators, cameras, displays, and Subscriber Identification Module (SIM) card interfaces, etc.
  • SIM Subscriber Identification Module
  • FIG. 1 it is a schematic diagram of a system architecture for performing a department triage method provided by an embodiment of the present application.
  • the system architecture mainly includes but is not limited to an information acquisition unit 100, a first processing unit 200.
  • Department prediction unit 300, second processing unit 400, and department recommendation unit 500 wherein:
  • the information acquisition unit 100 is used to acquire patient appeal information
  • the first processing unit 200 is configured to extract effective appeals from the appeal information through a deep learning entity recognition model, and perform feature extraction on the effective appeals, so as to extract features of effective appeals;
  • the department prediction unit 300 is used to input the effective appeal features into multiple different prediction models to perform department prediction respectively, and obtain the department prediction results output by each prediction model; wherein, the multiple prediction models include deep learning models, rule matching models, drug matching models, etc. model and disease matching model;
  • the second processing unit 400 is used to filter and sort the prediction results of all departments to obtain the sorting results
  • the department recommendation unit 500 is used to push target department information according to the sorting results.
  • each unit can call its stored program to implement the department triage method.
  • one embodiment of the present application provides a kind of department triage method, and this method comprises the following steps:
  • Step S100 acquiring patient appeal information.
  • the patient's appeal information can come from the terminal equipment on the Internet medical platform or the terminal equipment on the hospital treatment platform. said claim information).
  • a patient describes his symptoms, age, gender and other appeal information on the Internet medical platform. Help register."
  • the departmental triage method before step S100, also includes the following steps:
  • Step S1001 obtaining the historical consultation information of the patient
  • Step S1002 based on the patient's historical consultation information, push the target department information.
  • step S1002 is to push the target department information to the terminal equipment on the Internet medical platform or the terminal equipment on the hospital consultation platform. department or gastroenterology and so on.
  • Step S200 extract effective appeals from the appeal information by using the deep learning entity recognition model, and perform feature extraction on the effective appeals to obtain effective appeal features.
  • the effective appeal judgment of the patient's appeal information can eliminate invalid or defective information to ensure that useful features can be extracted, thereby ensuring the accuracy of the prediction results of multiple prediction departments obtained in subsequent steps.
  • judging the effective appeal of the patient's appeal information includes a processing process: extracting the effective appeal from the appeal information through the deep learning entity recognition model, and pushing the completion request of the appeal information (also known as is a questioning mechanism), when the completed appeal information is received, the effective appeal is extracted from the completed appeal information through the deep learning entity recognition model.
  • the patient's appeal information is: "Hello, what I want to consult is: My family member is not feeling well, and I want to take him to the hospital to see a doctor.” Which part of the body is uncomfortable”, “How long has the discomfort lasted”, “Whether you used drugs during the period of discomfort”, “Whether there is a history of disease”, etc.; if the patient's answer is: “It is my child who has nausea The situation, which appeared about 2 days ago, did not use drugs and had no medical history.”
  • effective appeals can be extracted through the deep learning entity recognition model. It should be noted that in step S200, the patient's appeal information is the corpus information belonging to the patient.
  • the corpus information often includes multiple entities (corresponding to the appeals described in the embodiments of this application), and the depth
  • the learning entity recognition model usually has an entity library, which contains many entity features. Through the deep learning entity recognition model and entity library, effective appeal judgment can be realized from the patient's appeal information.
  • the related technologies of the deep learning entity recognition model and the entity database are common knowledge, and the structure and principle of the deep learning entity recognition model will not be described in detail here. It is also worth noting that different entity types also correspond to fixed questioning words, which are not limited in this embodiment of the application, and can be set according to actual situations.
  • step S200 After the effective appeal is judged on the patient's appeal information in step S200, the features of the effective appeal are further extracted from the patient's appeal information.
  • the processing of extracting effective appeal features from patient appeal information includes, but is not limited to: segmenting patient appeal information, converting traditional Chinese to simplified Chinese, standardizing synonyms (that is, normalizing synonyms) and removing stop words etc., for example: the patient's appeal information put forward by the patient is: "Hello, doctor, I have a little stomachache.”;
  • the patient’s request information is: "Hello, what I want to consult is: I am pregnant and want to have a B-ultrasound examination, please register for me.” After processing, I get "[B-ultrasound] [wife] [pregnancy] [ registered ⁇ ”. It should be noted that the extracted effective appeal feature is a combination of two-dimensional word vectors.
  • Step S300 input the effective appeal features into multiple different prediction models to perform department prediction respectively, and obtain the department prediction results output by each prediction model, wherein the prediction models include a deep learning model, a rule matching model, a drug matching model and a disease matching model .
  • this step S300 Based on the effective appeal features obtained in the above step S200 as prediction data, this step S300 performs department predictions through a plurality of different prediction models, and each prediction model can correspond to output department prediction results.
  • the multiple prediction models in step S300 include: deep learning model, rule matching model, drug matching model and disease matching model, and each model is introduced as follows:
  • Step S301 input the effective appeal features into the trained deep learning model, and obtain the department prediction result output by the deep learning model.
  • the deep learning model includes Bayesian network model and BERT (Bidirectional Encoder Representations from Transformers) network model, compared with related schemes that use a single network model for training, this embodiment uses a combination of Bayesian network model and BERT network model for training and prediction schemes, which can avoid the limitations of a single model and increase the depth The accuracy of the predictions output by the learned model.
  • the Bayesian network model greatly considers the correlation characteristics of the disease, and has great advantages in knowledge reasoning.
  • the deep learning BERT model has made a great breakthrough in semantic reasoning, and has multiple triage results in prediction. has great advantages.
  • TF_IDF term frequency–inverse document frequency, a commonly used weighting scheme
  • the patient's age, gender, symptom description and other appeal information are directly input into the BERT network model for training.
  • Adjustable parameters include: learning rate, loss function, word vector dimension, number of iterations, batch data size (Batch Size) for updating gradients, etc.
  • the main method is to record and analyze the learning rate of each time step, the impact of loss rate changes on the accuracy of the model, and the impact of each training parameter on the accuracy of the model.
  • step S200 After obtaining the trained Bayesian network model and BERT network model, input the effective appeal features extracted in step S200 into the trained Bayesian network model and BERT network model, specifically including the following steps:
  • Step S3011 input the effective appeal features into the Bayesian network model and the BERT network model respectively.
  • Step S3012 when the output result of the Bayesian network model is the same as the output result of the BERT network model, use the same output result as the department prediction result.
  • Step S3013 when the output result of the Bayesian network model is different from the output result of the BERT network model, the output results of the Bayesian network model and the output results of the BERT network model are respectively subjected to confidence standardization processing to obtain a processing result; The processing results are sorted, and the top sorted output results are used as department prediction results.
  • Step S302 using the rule matching model to select departments that match the forward rules and do not match the reverse rules from the effective appeal features as department prediction results.
  • the rule matching model refers to the selection of departments that match the forward rules and do not match the reverse rules from the effective appeal features through the pre-configured regular expressions as the department prediction results.
  • each department corresponds to multiple forward and reverse rules. If the pre-configured regular expression matches the department’s forward rules and does not match the department’s reverse rules from the effective appeal features, Then the department is a department prediction result of the rule matching model.
  • Step S303 using the drug matching model to extract drug features from the effective appeal features, and match the department corresponding to the drug feature as the department prediction result.
  • Step S304 extracting disease features from the effective appeal features through the disease matching model, and matching the department corresponding to the disease feature as the department prediction result.
  • the drug matching model and the disease matching model use a similar method, that is, two data tables of drug characteristics and departments, disease characteristics and departments are obtained through data association analysis, because there is a certain relationship between drugs and departments, and the relationship between diseases and departments There is a certain correlation between them.
  • the drug feature is esomelaxol in the effective appeal feature, it can be seen that the drug is used to treat stomach diseases, and gastroenterology belongs to the department of gastroenterology. Therefore, the department prediction result output by the drug matching model is gastroenterology. The same is true for the disease matching model.
  • the patient's appeal information entered by the patient is: "Hello, what I want to consult is: the child has a fever and coughs all the time, but does not have a runny nose or dizziness”, then the extracted effective appeal
  • the features are: [Fever] [Always] [Cough] [Child].
  • the department prediction result output by the disease matching model based on the two disease characteristics of fever and cough is Respiratory Medicine.
  • Step S400 filtering and sorting the prediction results of all departments to obtain the sorting results.
  • step S400 specifically includes the following steps:
  • Step S401 Filter all department prediction results according to the patient's gender, age, preference, and historical consultation information to obtain filtered department prediction results.
  • the department prediction result of obstetrics and gynecology will be removed. If there are records of multiple consultations with traditional Chinese medicine in the patient's historical consultation information, then the Chinese medicine department label is given more weight, so that it is placed in the front position of multiple prediction results.
  • Step S402 sort the filtered department prediction results in sequence according to the order of the department prediction results output by the rule matching model, disease matching model, drug matching model and deep learning model.
  • the department recommended by the historical consultation information is given priority, and then the rule matching model, disease matching model, and drug matching model are sequentially used And the department prediction results output by the deep learning model are sorted and recommended. If there is no historical consultation information for the patient, the department prediction results output by the rule matching model, disease matching model, drug matching model, and deep learning model are used to sort and recommend.
  • Step S500 pushing target department information according to the sorting result.
  • the sorting results are sent to the terminal device on the Internet medical platform or the terminal device on the hospital consultation platform, so that the terminal device can recommend departments to patients according to the ranking results, and the department with the highest ranking is recommended first.
  • the department triage method also includes the steps of:
  • Step S601 acquiring referral information of the patient.
  • Step S602 iteratively updating the trained deep learning model according to the patient's referral information to obtain an updated deep learning model.
  • the terminal device After the department has been recommended to the patient in step S500, the terminal device will automatically match the doctor of the corresponding department. If the patient is not satisfied with the assignment of the department or the doctor thinks that the matching is wrong, the patient can actively choose to refer the patient or the doctor can choose to transfer the patient. After a referral occurs, referral information will be generated, and then step S601 will obtain the referral information as an error case for analysis and quality inspection (analysis and quality inspection can be performed by a doctor), and then the transfer after analysis and quality inspection
  • the diagnostic information is used as the training data of the Bayesian network model and the BERT network model, and then the Bayesian network model and the BERT network model are iterated and updated through the training data, so that the Bayesian network model and the BERT network model are gradually improved. , and finally improve the accuracy of the prediction results of the Bayesian network model and the BERT network model.
  • the patient’s appeal information is first obtained; then the effective appeal judgment is made on the patient’s appeal information, which can filter out unclear or invalid information caused by mistouching, thereby ensuring the accuracy of subsequent prediction results Accuracy; Then extract the effective appeal features, and based on the effective appeal features, perform department predictions through multiple different prediction models, and obtain multiple department prediction results.
  • the embodiment of this method can obtain multi-label prediction results, and can realize multi-label refined prediction, which not only improves the coverage of the prediction department, but also greatly improves the accuracy of the prediction department; Filtering and sorting the prediction results of each department can further improve the accuracy of the prediction department.
  • An embodiment of the present application provides an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and operable on the processor.
  • the processor and memory can be connected by a bus or other means.
  • memory can be used to store non-transitory software programs and non-transitory computer-executable programs.
  • the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage devices.
  • the memory optionally includes memory located remotely from the processor, and these remote memories may be connected to the processor via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
  • the electronic device in this embodiment can constitute a part of the system architecture in the embodiment shown in FIG. 1, and these embodiments all belong to the same inventive concept, so these embodiments have the same implementation principle and technical effect, No more details here.
  • the non-transitory software programs and instructions required to realize the department triage method of the above-mentioned embodiment are stored in the memory, and when executed by the processor, the method of the above-mentioned embodiment is executed, for example, the method step S100 in Fig. 3 described above is executed to S500.
  • terminal embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • an embodiment of the embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium may be non-volatile or volatile, and the computer-readable storage medium stores the computer Executable instructions, the computer-executable instructions are executed by a processor or a controller, for example, executed by a processor in the above-mentioned electronic device embodiment, so that the above-mentioned processor can execute the department triage method in the above-mentioned embodiment, for example , executing the method steps S100 to S500 in FIG. 3 described above.
  • being executed by a processor in the above embodiment of the device connector may cause the above processor to execute the department triage method in the above embodiment, for example, execute the method steps S100 to S500 in FIG. 3 described above.
  • Computer storage media including, but not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cartridges, tape, magnetic disk storage or other magnetic storage devices, or Any other medium used to store desired information and which can be accessed by a computer.
  • communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media .

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Business, Economics & Management (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Biomedical Technology (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Data Mining & Analysis (AREA)
  • Economics (AREA)
  • Databases & Information Systems (AREA)
  • Operations Research (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Quality & Reliability (AREA)
  • Marketing (AREA)
  • Epidemiology (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Medical Informatics (AREA)
  • Primary Health Care (AREA)
  • Public Health (AREA)
  • Tourism & Hospitality (AREA)
  • Game Theory and Decision Science (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Development Economics (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

提供一种科室分诊方法、系统、设备以及存储介质，涉及智慧医疗及人工智能领域，首先获取患者的诉求信息；然后对患者的诉求信息进行有效诉求判断，能够过滤掉不清楚或者因误触而产生的无效信息，从而确保后续预测结果的准确度；然后再提取有效诉求特征，并基于有效诉求特征，通过深度学习模型、规则匹配模型、药品匹配模型以及疾病匹配模型分别进行预测，得到每个预测模型预测出的科室预测结果，相较于相关方案的预测结果为单标签的预测结果，能够得到多标签的预测结果，提升了预测科室的覆盖率和准确度；最后还对预测出的全部科室预测结果进行过滤和排序，能够进一步的提升预测科室的准确度。

Description

科室分诊方法、系统、设备以及存储介质
本申请要求于2021年08月30日提交中国专利局、申请号为202111004639.3，发明名称为“科室分诊方法、系统、设备以及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请实施例涉及智慧医疗技术领域，尤其涉及一种科室分诊方法、系统、设备以及存储介质。
 背景技术
分诊是医学人工智能和互联网医疗中研究的领域，也是病患就诊时的入口路径，分诊错误会导致后续一系列问诊流程出错的连锁反应。
传统的分诊通常需要病患自己对身体不适的判断主动挂号对应科室问诊，这就造成了相当一部分的患者因为对医疗知识的欠缺，而挂号到错误科室，不仅耽误就诊时间又耗费精力。现有多采用智能分诊技术，智能分诊通常是指医院或互联网医疗机构在征得患者同意的情况下，收集病患的描述的主诉（例如性别、年龄以及对自己身体异常的描述信息等），接着医疗从业人员对主诉进行专业分析后，将该主诉标注到对应的科室，当收集到一定量的标注数据后，使用深度学习算法训练预测模型，最后基于训练完成之后的模型对病患的主诉进行科室的预测，但发明人意识到该方案的预测结果多为单标签结果，不仅预测科室的覆盖率较低，而且预测科室的准确度也不高。而且患者提供的主诉有大量的表述不清或者误触输入的情况，在相关方案中，会忽略主诉中存在的不清楚或误触的情况，从而降低后续的预测结果的准确度。
 技术问题
以下是对本文详细描述的主题的概述。本概述并非是为了限制权利要求的保护范围。
本申请实施例提供了一种科室分诊方法、系统、设备以及存储介质，科室预测结果的覆盖度更高，而且科室推荐的准确度也更高。
 技术解决方案
第一方面，本申请实施例提供一种科室分诊方法，所述科室分诊方法包括：
获取患者的诉求信息；
通过深度学习实体识别模型从所述诉求信息中提取出有效诉求，并对所述有效诉求进行特征提取，以提取得到有效诉求特征；
将所述有效诉求特征输入至多个不同的预测模型分别进行科室预测，得到每个所述预测模型输出的科室预测结果；其中，所述预测模型包括深度学习模型、规则匹配模型、药品匹配模型以及疾病匹配模型；
对全部所述科室预测结果进行过滤和排序，得到排序结果；
根据所述排序结果推送目标科室信息。
第二方面，本申请实施例提供一种科室分诊系统，所述科室分诊系统包括：
信息获取单元，用于获取患者的诉求信息；
第一处理单元，用于通过深度学习实体识别模型从所述诉求信息中提取出有效诉求，并对所述有效诉求进行特征提取，以提取得到有效诉求特征；
科室预测单元，用于将所述有效诉求特征输入至多个不同的预测模型分别进行科室预测，得到每个所述预测模型输出的科室预测结果；其中，所述预测模型包括深度学习模型、规则匹配模型、药品匹配模型以及疾病匹配模型；
第二处理单元，用于对全部所述科室预测结果进行过滤和排序，得到排序结果；
科室推荐单元，用于根据所述排序结果推送目标科室信息。
第三方面，本申请实施例提供一种电子设备，包括：存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现：如上述的科室分诊方法。
第四方面，本申请实施例提供一种计算机可读存储介质，存储有计算机可执行指令，所述计算机可执行指令用于执行：如上述的科室分诊方法。
 有益效果
本申请实施例第一方面提供的一种科室分诊方法，本方法首先获取患者的诉求信息；然后对患者的诉求信息进行有效诉求判断，能够过滤掉不清楚或者因误触而产生的无效信息，从而确保后续预测结果的准确度；然后再提取有效诉求特征，并基于该有效诉求特征，通过深度学习模型、规则匹配模型、药品匹配模型以及疾病匹配模型分别进行预测分别进行预测，得到每个预测模型预测出的科室预测结果，相较于相关方案的预测结果为单标签的预测结果，本方法通过深度学习模型、规则匹配模型、药品匹配模型以及疾病匹配模型分别进行预测，能够得到多标签的预测结果，不仅提升了预测科室的覆盖率，而且提升了预测科室的准确度；最后本方法还对预测出的全部科室预测结果进行过滤和排序，能够进一步的提升预测科室的准确度。
可以理解的是，上述第二方面至第四方面与相关技术相比存在的有益效果与上述第一方面与相关技术相比存在的有益效果相同，可以参见上述第一方面中的相关描述，在此不再赘述。
 附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例或相关技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请实施例的一些实施例，对于本领域普通技术人员来说，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1是本申请一个实施例提供的用于执行科室分诊方法的系统架构的示意图；
图2是本申请一个实施例提供的一种科室分诊方法的逻辑框图；
图3是本申请一个实施例提供的一种科室分诊方法的流程示意图。
 本发明的实施方式
以下描述中，为了说明而不是为了限定，提出了诸如特定系统结构、技术之类的具体细节，以便透彻理解本申请实施例。然而，本领域的技术人员应当清楚，在没有这些具体细节的其它实施例中也可以实现本申请实施例。在其它情况中，省略对众所周知的系统、装置、电路以及方法的详细说明，以免不必要的细节妨碍本申请实施例的描述。
需要说明的是，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于流程图中的顺序执行所示出或描述的步骤。说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。
还应当理解，在本申请实施例说明书中描述的参考“一个实施例”或“一些实施例”等意味着在本申请实施例的一个或多个实施例中包括结合该实施例描述的特定特征、结构或特点。由此，在本说明书中的不同之处出现的语句“在一个实施例中”、“在一些实施例中”、“在其他一些实施例中”、“在另外一些实施例中”等不是必然都参考相同的实施例，而是意味着“一个或多个但不是所有的实施例”，除非是以其他方式另外特别强调。术语“包括”、“包含”、“具有”及它们的变形都意味着“包括但不限于”，除非是以其他方式另外特别强调。
本申请实施例可以基于医疗人工智能技术对相关的数据进行获取和处理。其中，人工智能（Artificial-intelligence，AI）是利用数字计算机或者数字计算机控制的机器模拟、延伸和扩展人的智能，感知环境、获取知识并使用知识获得最佳结果的理论、方法、技术及应用系统。人工智能基础技术一般包括如传感器、专用人工智能芯片、云计算、分布式存储、大数据处理技术、操作/交互系统、机电一体化等技术。人工智能软件技术主要包括计算机视觉技术、机器人技术、生物识别技术、语音处理技术、自然语言处理技术以及机器学习/深度学习等几大方向。本申请实施例主要涉及医疗人工智能中的自然语言处理技术和机器学习/深度学习技术。
分诊是医学人工智能和互联网医疗中研究的领域，也是病患就诊时的入口路径，分诊错误会导致后续一系列问诊流程出错的连锁反应。
传统的分诊通常需要病患自己对身体不适的判断主动挂号对应科室问诊，这就造成了相当一部分的患者因为对医疗知识的欠缺，而挂号到错误科室，不仅耽误就诊时间又耗费精力。现有多采用智能分诊技术，智能分诊通常是指医院或互联网医疗机构在征得患者同意的情况下，收集病患的描述的主诉（例如性别、年龄以及对自己身体异常的描述信息等），接着医疗从业人员对主诉进行专业分析后，将该主诉标注到对应的科室，当收集到一定量的标注数据后，使用深度学习算法训练预测模型，最后基于训练完成之后的模型对病患的主诉进行科室的预测，但该方案的预测结果多为单标签结果，不仅预测科室的覆盖率较低，而且预测科室的准确度也不高。而且患者提供的主诉有大量的表述不清或者误触输入的情况，在相关方案中，会忽略主诉中存在的不清楚或误触的情况，从而降低后续的预测结果的准确度。
为了解决上述技术缺陷，本申请实施例首先获取患者的诉求信息；然后对患者的诉求信息进行有效诉求判断，能够过滤掉不清楚或者因误触而产生的无效信息，从而确保后续预测结果的准确度；然后再提取有效诉求特征，并基于该有效诉求特征，通过多个不同的预测模型分别进行科室预测，得到多个科室预测结果，相较于相关方案的预测结果为单标签的预测结果，本本申请实施例能够得到多标签的预测结果，能够实现多标签精细化预测，不仅提升了预测科室的覆盖率，而且也极大的提升了预测科室的准确度；最后本申请实施例还对多个科室预测结果进行过滤和排序，能够进一步的提升预测科室的准确度。
本申请一个实施例提供的科室分诊方法可以在电子设备中执行。终端/设备可以为移动电子设备，也可以为非移动电子设备。移动电子设备可以为手机、平板电脑、笔记本电脑、掌上电脑、车载电子设备、可穿戴设备、超级移动个人计算机、上网本、个人数字助理等；非移动电子设备可以为个人计算机、电视机、柜员机或者自助机等；本申请实施方案不作具体限定。
电子设备可以包括处理器，外部存储器接口，内部存储器，通用串行总线(universal serial bus，USB)接口，充电管理模块，电源管理模块，电池，天线，移动通信模块，无线通信模块，音频模块，扬声器，受话器，麦克风，耳机接口，传感器模块，按键，马达，指示器，摄像头，显示屏，以及用户标识模块(Subscriber Identification Module，SIM)卡接口等。
下面结合附图，对本申请实施例作进一步阐述。
参照图1，是本申请一个实施例提供的用于执行一种科室分诊方法的系统架构示意图，在图1的示例中，该系统架构主要包括但不限于信息获取单元100、第一处理单元200、科室预测单元300、第二处理单元400以及科室推荐单元500，其中：
信息获取单元100用于获取患者的诉求信息；
第一处理单元200用于通过深度学习实体识别模型从诉求信息中提取出有效诉求，并对有效诉求进行特征提取，以提取得到有效诉求特征；
科室预测单元300用于将有效诉求特征输入至多个不同的预测模型分别进行科室预测，得到每个预测模型输出的科室预测结果；其中，多个预测模型包括深度学习模型、规则匹配模型、药品匹配模型以及疾病匹配模型；
第二处理单元400用于对全部科室预测结果进行过滤和排序，得到排序结果；
科室推荐单元500用于根据排序结果推送目标科室信息。
本申请实施例描述的系统架构以及应用场景是为了更加清楚的说明本申请实施例的技术方案，并不构成对于本申请实施例提供的技术方案的限定，本领域技术人员可知，随着系统架构的演变和新应用场景的出现，本申请实施例提供的技术方案对于类似的技术问题，同样适用。
本领域技术人员可以理解的是，图1中示出的系统架构并不构成对本申请实施例的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。
在图1所示的系统架构中，各个单元可以分别调用其储存的程序，用以执行科室分诊方法。
基于上述系统架构，提出本申请实施例的科室分诊方法的各个实施例。
参照图2，本申请的一个实施例，提供了一种科室分诊方法，本方法包括如下步骤：
步骤S100、获取患者的诉求信息。
这里，患者的诉求信息可以来自于互联网医疗平台的终端设备或医院就诊平台上的终端设备，例如患者在互联网医疗平台或者医院进行就诊时，将先于终端设备上提出诉求（即本申请实施例所述诉求信息）。例如某患者在互联网医疗平台上描述自己症状、年龄、性别等诉求信息，其所提出的诉求信息为：“你好，我想咨询的是：我老婆有身孕了，想做B超检查，请帮忙挂号”。
在一些实施例中，在步骤S100之前，本科室分诊方法还包括如下步骤：
步骤S1001、获取患者的历史问诊信息；
步骤S1002、基于患者的历史问诊信息，推送目标科室信息。
由于患者的历史问诊信息是一项对患者重要的参考数据（在实际应用中，医生也会根据患者的历史问诊信息来辅助诊断），本实施例在对患者进行多个预测模型的科室分诊预测之前，先判断患者是否存在历史问诊信息，当患者有历史问诊信息的时候，可先在互联网医疗平台或者医院就诊平台中获取历史问诊信息，历史问诊信息中包括患者进行问诊科室的相关数据，然后再依据历史问诊信息对患者进行相应科室推荐，例如患者的历史问诊信息中多次记录其咨询中医，则优先向患者推荐中医科室。若患者不选择中医科室，则继续后续的步骤。需要注意的是，步骤S1002是向互联网医疗平台的终端设备或医院就诊平台上的终端设备推送目标科室信息，本申请所述目标科室信息中包含有向患者推荐的科室数据，例如向患者推荐中医科室或消化内科等等。
步骤S200、通过深度学习实体识别模型从诉求信息中提取出有效诉求，并对有效诉求进行特征提取，以提取得到有效诉求特征。
在本实施例中，首先对患者的诉求信息进行有效诉求判断能剔除无效或存在缺陷的信息，确保能提取出有用的特征，从而确保后续步骤得到的多个预测科室预测结果的准确度。在一些实施例中，对患者的诉求信息进行有效诉求判断包括处理过程：通过深度学习实体识别模型从诉求信息中提取有效诉求，当未提取到有效诉求，推送诉求信息的补全请求（又称为追问机制），当接收到补全后的诉求信息，通过深度学习实体识别模型从补全后的诉求信息中提取出有效诉求。例如患者的诉求信息为：“你好，我想咨询的是：我家人身体不舒服，我想带他来医院看医生”，追问包括但不仅限于：“请问是哪位家人”、“请问是身体哪个部位不舒服”、“请问不舒服已持续多长时间”、“请问在不舒服期间是否使用药物”、“请问是否存在疾病史”等；如患者的回答为：“是我家小孩出现反胃的情况，大约是2天前出现的这种情况，没有使用药物，也没有疾病史”。在得到患者补全之后的诉求信息，就可通过深度学习实体识别模型提取出有效诉求。需要注意的是，在步骤S200当中，患者的诉求信息是属于患者的语料信息，在实际应用中，语料信息中往往包括多个实体（对应于本申请实施例中所述的诉求），而且深度学习实体识别模型通常具备一个实体库，实体库中包含着众多的实体特征，通过深度学习实体识别模型和实体库可以实现从患者的诉求信息中进行有效诉求判断。深度学习实体识别模型和实体库的相关技术均属于公知常识，此处便不再细述该深度学习实体识别模型的结构和原理。还值得注意的是，不同的实体类型还对应着固定的追问话术，本申请实施例不作任何限制，可以根据现实情况进行设置。
当步骤S200对患者的诉求信息进行有效诉求判断后，进一步的从患者的诉求信息中提取出有效诉求特征。
在一些实施例中，对从患者的诉求信息中提取出有效诉求特征的处理包括但不仅限于：对患者的诉求信息进行分词、繁体转简体、同义词标准化（即同义词归一）以及去除停用词等处理，例如：患者提出的患者的诉求信息为：“您好，醫生，我我有點肚子痛。”；
那么繁转简后为：“您好，医生，我我有点肚子痛。”；
同义词标准化之后为：“您好，医生，我有点肚子痛。”；
分词后为：“【您好】【，】【医生】【，】【我】【有点】【肚子痛】【。】”；
去掉停用词后为：“【我】【有点】【肚子痛】”。
例如：患者提出诉求信息为：“你好，我想咨询的是：我有身孕了，想做B超检查，请帮忙挂号”，进行处理之后得到“【b超】【妻子】【怀孕】【挂号】”。需要注意的是，提取的有效诉求特征为二维词向量组合。
步骤S300、将有效诉求特征输入至多个不同的预测模型分别进行科室预测，得到每个预测模型输出的科室预测结果，其中，预测模型包括深度学习模型、规则匹配模型、药品匹配模型以及疾病匹配模型。
基于上述步骤S200求取的有效诉求特征作为预测数据，本步骤S300通过多个不同的预测模型分别进行科室预测，能够每个预测模型对应输出的科室预测结果。步骤S300中的多个预测模型包括：深度学习模型、规则匹配模型、药品匹配模型和疾病匹配模型，以下分别对每一个模型进行介绍：
步骤S301、将有效诉求特征输入已训练完成的深度学习模型中，得到深度学习模型输出的科室预测结果。
在一些实施例中，深度学习模型包括贝叶斯网络模型和BERT（Bidirectional Encoder Representations from Transformers）网络模型，相较于相关方案使用单个网络模型进行训练，本实施例使用贝叶斯网络模型和BERT网络模型结合的训练和预测方案，能够避免单个模型的局限性，增加深度学习模型输出的预测结果的准确度。而且贝叶斯网络模型极大的考虑病情关联特征，在知识推理方面具有较大的优势，深度学习BERT模型在语义推理上有了极大的突破，并且在具有多个分诊结果上预测上具有很大的优势。
以下简要介绍贝叶斯网络模型和BERT网络模型的训练过程：
在使用贝叶斯网络模型的时候需要分词，去除停用词，计算TF_IDF（term frequency–inverse document frequency，一种常用的加权方案），提取有效的特征，然后把特征输入给贝叶斯网络模型训练。特征处理具体为：首先将输入文本中的英文统一为小写，通过预先构造的停用词表进行过滤，再分词，而后将分词的内容进行同义词标准化。
在使用BERT网络模型的时候，把患者的年龄、性别、症状描述等诉求信息，直接输入给BERT网络模型训练。
训练时会进行多次调参过程，用于得到一个相对表现较好的模型。可调参数包括：学习率、损失函数、词向量维度、迭代次数、更新梯度的批数据大小（Batch Size）等。主要方法为记录和分析每个时间步骤的学习率、损失率变化对模型准确率的影响，以及每次训练参数对模型准确率的影响。
在得到已经训练完成的贝叶斯网络模型和BERT网络模型之后，将步骤S200提取出的有效诉求特征分别输入至已经训练完成的贝叶斯网络模型和BERT网络模型中，具体包括如下步骤：
步骤S3011、将有效诉求特征分别输入至贝叶斯网络模型和BERT网络模型。
步骤S3012、当贝叶斯网络模型的输出结果与BERT网络模型的输出结果相同时，将相同的输出结果作为科室预测结果。
步骤S3013、当贝叶斯网络模型的输出结果与BERT网络模型的输出结果不相同时，将贝叶斯网络模型的输出结果和BERT网络模型的输出结果分别进行置信度标准化处理，得到处理结果；将处理结果进行排序，将排序最前的输出结果作为科室预测结果。
步骤S302、通过规则匹配模型从有效诉求特征中选取匹配出正向规则且没有匹配出逆向规则的科室作为科室预测结果。
规则匹配模型是指的通过预先配置的正则表达式从有效诉求特征中选取匹配出正向规则且没有匹配出逆向规则的科室作为科室预测结果。在规则匹配模型中，每个科室对应有多条正向和逆向的规则，如果通过预先配置的正则表达式从有效诉求特征中匹配到科室的正向规则且未匹配到该科室的逆向规则，则该科室为规则匹配模型的一个科室预测结果。
步骤S303、通过药品匹配模型从有效诉求特征中提取药品特征，匹配与药品特征对应的科室作为科室预测结果。
步骤S304、通过疾病匹配模型从有效诉求特征中提取疾病特征，匹配与疾病特征对应的科室作为科室预测结果。
药品匹配模型和疾病匹配模型是使用类似的方法，即通过数据关联分析得到药品特征与科室、疾病特征与科室两个数据表，因为药品与科室之间存在着一定的关联关系，疾病与科室之间存在着一定的关联关系，例如有效诉求特征出现药品特征为埃索美拉素，可知该药品是用于治疗胃病，而胃病属于消化内科，因此药品匹配模型输出的科室预测结果为消化内科。疾病匹配模型也是同理，例如：患者输入的患者的诉求信息为：“你好，我想咨询的是：孩子发热，一直咳嗽，但没有流鼻涕或头晕的情况”，那么提取出的有效诉求特征为：【发热】【总是】【咳嗽】【孩子】，最后疾病匹配模型基于发热和咳嗽这两个疾病特征输出的科室预测结果为呼吸内科。
步骤S400、对全部科室预测结果进行过滤和排序，得到排序结果。
在步骤S300中得到了四个预测模型输出的多标签结果之后，需要进一步对多个科室预测结果进行过滤和排序。在一些实施例中，步骤S400具体包括如下步骤：
步骤S401、根据患者的性别、年龄、偏好以及历史问诊信息对全部科室预测结果进行过滤，得到过滤后的科室预测结果。
例如：若患者的性别为男性，而科室预测结果中出现妇产科，则去掉妇产科的科室预测结果。若患者的历史问诊信息当中存在有多次咨询中医的记录，那么将中医科标签赋予更多权重，使其置于多个预测结果的靠前位置。
步骤S402、依次按照规则匹配模型、疾病匹配模型、药品匹配模型以及深度学习模型输出的科室预测结果的顺序对过滤后的科室预测结果进行排序。
在本实施例中，如果患者存在历史问诊信息（如上述步骤S1001和步骤S1002所示），以历史问诊信息推荐的科室作为优先，然后依次以规则匹配模型、疾病匹配模型、药品匹配模型以及深度学习模型输出的科室预测结果进行排序并进行推荐。若患者不存在历史问诊信息，依次以规则匹配模型、疾病匹配模型、药品匹配模型以及深度学习模型输出的科室预测结果进行排序并进行推荐。
步骤S500、根据排序结果推送目标科室信息。
在本申请实施例之中，将排序结果发送至互联网医疗平台的终端设备或医院就诊平台上的终端设备，以使终端设备能够依据排序结果对患者进行科室推荐，先推荐排序最前的科室。
在一些实施例，在步骤S500之后，本科室分诊方法还包括步骤：
步骤S601、获取患者的转诊信息。
步骤S602、通过患者的转诊信息对已训练完成的深度学习模型进行迭代更新，得到更新后的深度学习模型。
在步骤S500已经向患者推荐科室之后，终端设备将会自动匹配对应科室的医生，若患者不满意科室的分配或医生认为匹配有误，则患者可以主动选择转诊或医生可以选择将患者进行转诊，在发生转诊之后，会产生转诊信息，然后步骤S601将获取该转诊信息作为错误案件进行分析和质检（可由医生进行分析和质检），然后将分析和质检之后的转诊信息作为贝叶斯网络模型和BERT网络模型的训练数据，然后通过该训练数据对贝叶斯网络模型和BERT网络模型进行模型迭代和更新，从而使贝叶斯网络模型和BERT网络模型逐步完善，最终提高贝叶斯网络模型和BERT网络模型科室预测结果的准确度。
本申请提供的科室分诊方法实施例，首先获取患者的诉求信息；然后对患者的诉求信息进行有效诉求判断，能够过滤掉不清楚或者因误触而产生的无效信息，从而确保后续预测结果的准确度；然后再提取有效诉求特征，并基于该有效诉求特征，通过多个不同的预测模型分别进行科室预测，得到多个科室预测结果，相较于相关方案的预测结果为单标签的预测结果，本方法实施例能够得到多标签的预测结果，能够实现多标签精细化预测，不仅提升了预测科室的覆盖率，而且也极大的提升了预测科室的准确度；最后本实施例还对多个科室预测结果进行过滤和排序，能够进一步的提升预测科室的准确度。
本申请的一个实施例，提供了一种电子设备，该设备包括：存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序。
处理器和存储器可以通过总线或者其他方式连接。
存储器作为一种非暂态计算机可读存储介质，可用于存储非暂态软件程序以及非暂态性计算机可执行程序。此外，存储器可以包括高速随机存取存储器，还可以包括非暂态存储器，例如至少一个磁盘存储器件、闪存器件、或其他非暂态固态存储器件。在一些实施方式中，存储器可选包括相对于处理器远程设置的存储器，这些远程存储器可以通过网络连接至该处理器。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
需要说明的是，本实施例中的电子设备能够构成图1所示实施例中的系统架构的一部分，这些实施例均属于相同的发明构思，因此这些实施例具有相同的实现原理以及技术效果，此处不再详述。
实现上述实施例的科室分诊方法所需的非暂态软件程序以及指令存储在存储器中，当被处理器执行时，执行上述实施例方法，例如，执行以上描述的图3中的方法步骤S100至S500。
以上所描述的终端实施例仅仅是示意性的，其中作为分离部件说明的单元可以是或者也可以不是物理上分开的，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。
此外，本申请实施例的一个实施例还提供了一种计算机可读存储介质，所述计算机可读存储介质可以是非易失性，也可以是易失性，该计算机可读存储介质存储有计算机可执行指令，该计算机可执行指令被一个处理器或控制器执行，例如，被上述电子设备实施例中的一个处理器执行，可使得上述处理器执行上述实施例中的科室分诊方法，例如，执行以上描述的图3中的方法步骤S100至S500。
又如，被上述设备连接器实施例中的一个处理器执行，可使得上述处理器执行上述实施例中的科室分诊方法，例如，执行以上描述的图3中的方法步骤S100至S500。
本领域普通技术人员可以理解，上文中所公开方法中的全部或某些步骤、系统可以被实施为软件、固件、硬件及其适当的组合。某些物理组件或所有物理组件可以被实施为由处理器，如中央处理器、数字信号处理器或微处理器执行的软件，或者被实施为硬件，或者被实施为集成电路，如专用集成电路。这样的软件可以分布在计算机可读介质上，计算机可读介质可以包括计算机存储介质（或非暂时性介质）和通信介质（或暂时性介质）。如本领域普通技术人员公知的，术语计算机存储介质包括在用于存储信息（诸如计算机可读指令、数据结构、程序模块或其他数据）的任何方法或技术中实施的易失性和非易失性、可移除和不可移除介质。计算机存储介质包括但不限于RAM、ROM、EEPROM、闪存或其他存储器技术、CD-ROM、数字多功能盘（DVD）或其他光盘存储、磁盒、磁带、磁盘存储或其他磁存储装置、或者可以用于存储期望的信息并且可以被计算机访问的任何其他的介质。此外，本领域普通技术人员公知的是，通信介质通常包含计算机可读指令、数据结构、程序模块或者诸如载波或其他传输机制之类的调制数据信号中的其他数据，并且可包括任何信息递送介质。
以上是对本申请实施例的较佳实施进行了具体说明，但本申请实施例并不局限于上述实施方式，熟悉本领域的技术人员在不违背本申请实施例精神的前提下还可作出种种的等同变形或替换，这些等同的变形或替换均包含在本申请实施例权利要求所限定的范围内。

Claims (20)

  1. 一种科室分诊方法，其中，所述科室分诊方法包括：
    获取患者的诉求信息；
    通过深度学习实体识别模型从所述诉求信息中提取出有效诉求，并对所述有效诉求进行特征提取，以提取得到有效诉求特征；
    将所述有效诉求特征输入至多个不同的预测模型分别进行科室预测，得到每个所述预测模型输出的科室预测结果；其中，所述预测模型包括深度学习模型、规则匹配模型、药品匹配模型以及疾病匹配模型；
    对全部所述科室预测结果进行过滤和排序，得到排序结果；
    根据所述排序结果推送目标科室信息。
  2. 根据权利要求1所述的科室分诊方法，其中，所述将所述有效诉求特征输入至多个不同的预测模型分别进行科室预测，得到每个所述预测模型输出的科室预测结果，包括：
    将所述有效诉求特征输入已训练完成的所述深度学习模型中，得到所述深度学习模型输出的科室预测结果；
    将所述有效诉求特征输入至所述规则匹配模型中，得到所述规则匹配模型输出的科室预测结果；其中所述规则匹配模型是根据所述有效诉求特征，将匹配出正向规则且没有匹配出逆向规则的科室作为科室预测结果；
    将所述有效诉求特征输入至所述药品匹配模型中，得到所述药品匹配模型输出的科室预测结果；其中所述药品匹配模型是根据所述有效诉求特征获取对应的药品特征，将匹配与所述药品特征对应的科室作为科室预测结果；
    将所述有效诉求特征输入至所述疾病匹配模型中，得到所述疾病匹配模型输出的科室预测结果；其中所述疾病匹配模型是根据所述有效诉求特征获取对应的疾病特征，将匹配与所述疾病特征对应的科室作为科室预测结果。
  3. 根据权利要求2所述的科室分诊方法，其中，所述深度学习模型包括已训练完成的贝叶斯网络模型和BERT网络模型；所述将所述有效诉求特征输入已训练完成的所述深度学习模型中，得到所述深度学习模型输出的科室预测结果，包括：
    将所述有效诉求特征分别输入至所述贝叶斯网络模型和所述BERT网络模型；
    当所述贝叶斯网络模型的输出结果与所述BERT网络模型的输出结果相同时，将相同的输出结果作为科室预测结果；
    当所述贝叶斯网络模型的输出结果与所述BERT网络模型的输出结果不相同时，将所述贝叶斯网络模型的输出结果和所述BERT网络模型的输出结果分别进行置信度标准化处理，得到处理结果；
    将所述处理结果进行排序，将排序最前的输出结果作为科室预测结果。
  4. 根据权利要求2所述的科室分诊方法，其中，所述对全部所述科室预测结果进行过滤和排序，得到排序结果，包括：
    根据所述患者的性别、年龄、偏好以及历史问诊信息对全部所述科室预测结果进行过滤，得到过滤后的所述科室预测结果；
    依次按照所述规则匹配模型、所述疾病匹配模型、所述药品匹配模型以及所述深度学习模型输出的科室预测结果的顺序对所述过滤后的所述科室预测结果进行排序。
  5. 根据权利要求1所述的科室分诊方法，其中，所述科室分诊方法还包括：
    当通过所述深度学习实体识别模型从所述诉求信息中没有提取到有效诉求，推送所述诉求信息的补全请求；
    当接收到补全后的诉求信息，通过所述深度学习实体识别模型从所述补全后的诉求信息中提取出有效诉求。
  6. 根据权利要求2所述的科室分诊方法，其中，在所述根据所述排序结果推送目标科室信息之后，所述科室分诊方法还包括：
    获取患者的转诊信息；
    通过所述转诊信息对已训练完成的所述深度学习模型进行迭代更新，得到更新后的所述深度学习模型。
  7. 根据权利要求1所述的科室分诊方法，其中，在所述获取患者的诉求信息之前，所述科室分诊方法还包括：
    获取患者的历史问诊信息；
    根据所述患者的历史问诊信息，推送目标科室信息。
  8. 一种科室分诊系统，其中，所述科室分诊系统包括：
    信息获取单元，用于获取患者的诉求信息；
    第一处理单元，用于通过深度学习实体识别模型从所述诉求信息中提取出有效诉求，并对所述有效诉求进行特征提取，以提取得到有效诉求特征；
    科室预测单元，用于将所述有效诉求特征输入至多个不同的预测模型分别进行科室预测，得到每个所述预测模型输出的科室预测结果；其中，所述预测模型包括深度学习模型、规则匹配模型、药品匹配模型以及疾病匹配模型；
    第二处理单元，用于对全部所述科室预测结果进行过滤和排序，得到排序结果；
    科室推荐单元，用于根据所述排序结果推送目标科室信息。
  9. 一种电子设备，包括：存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其中，所述处理器执行所述计算机程序时实现一项所述的科室分诊方法，所述科室分诊方法包括：
    获取患者的诉求信息；
    通过深度学习实体识别模型从所述诉求信息中提取出有效诉求，并对所述有效诉求进行特征提取，以提取得到有效诉求特征；
    将所述有效诉求特征输入至多个不同的预测模型分别进行科室预测，得到每个所述预测模型输出的科室预测结果；其中，所述预测模型包括深度学习模型、规则匹配模型、药品匹配模型以及疾病匹配模型；
    对全部所述科室预测结果进行过滤和排序，得到排序结果；
    根据所述排序结果推送目标科室信息。
  10. 根据权利要求9所述的电子设备，其中，所述将所述有效诉求特征输入至多个不同的预测模型分别进行科室预测，得到每个所述预测模型输出的科室预测结果，包括：
    将所述有效诉求特征输入已训练完成的所述深度学习模型中，得到所述深度学习模型输出的科室预测结果；
    将所述有效诉求特征输入至所述规则匹配模型中，得到所述规则匹配模型输出的科室预测结果；其中所述规则匹配模型是根据所述有效诉求特征，将匹配出正向规则且没有匹配出逆向规则的科室作为科室预测结果；
    将所述有效诉求特征输入至所述药品匹配模型中，得到所述药品匹配模型输出的科室预测结果；其中所述药品匹配模型是根据所述有效诉求特征获取对应的药品特征，将匹配与所述药品特征对应的科室作为科室预测结果；
    将所述有效诉求特征输入至所述疾病匹配模型中，得到所述疾病匹配模型输出的科室预测结果；其中所述疾病匹配模型是根据所述有效诉求特征获取对应的疾病特征，将匹配与所述疾病特征对应的科室作为科室预测结果。
  11. 根据权利要求10所述的电子设备，其中，所述深度学习模型包括已训练完成的贝叶斯网络模型和BERT网络模型；所述将所述有效诉求特征输入已训练完成的所述深度学习模型中，得到所述深度学习模型输出的科室预测结果，包括：
    将所述有效诉求特征分别输入至所述贝叶斯网络模型和所述BERT网络模型；
    当所述贝叶斯网络模型的输出结果与所述BERT网络模型的输出结果相同时，将相同的输出结果作为科室预测结果；
    当所述贝叶斯网络模型的输出结果与所述BERT网络模型的输出结果不相同时，将所述贝叶斯网络模型的输出结果和所述BERT网络模型的输出结果分别进行置信度标准化处理，得到处理结果；
    将所述处理结果进行排序，将排序最前的输出结果作为科室预测结果。
  12. 根据权利要求10所述的电子设备，其中，所述对全部所述科室预测结果进行过滤和排序，得到排序结果，包括：
    根据所述患者的性别、年龄、偏好以及历史问诊信息对全部所述科室预测结果进行过滤，得到过滤后的所述科室预测结果；
    依次按照所述规则匹配模型、所述疾病匹配模型、所述药品匹配模型以及所述深度学习模型输出的科室预测结果的顺序对所述过滤后的所述科室预测结果进行排序。
  13. 根据权利要求9所述的电子设备，其中，所述科室分诊方法还包括：
    当通过所述深度学习实体识别模型从所述诉求信息中没有提取到有效诉求，推送所述诉求信息的补全请求；
    当接收到补全后的诉求信息，通过所述深度学习实体识别模型从所述补全后的诉求信息中提取出有效诉求。
  14. 根据权利要求10所述的电子设备，其中，在所述根据所述排序结果推送目标科室信息之后，所述科室分诊方法还包括：
    获取患者的转诊信息；
    通过所述转诊信息对已训练完成的所述深度学习模型进行迭代更新，得到更新后的所述深度学习模型。
  15. 一种计算机可读存储介质，存储有计算机可执行指令，其中，所述计算机可执行指令用于执行一种科室分诊方法，所述科室分诊方法包括：
    获取患者的诉求信息；
    通过深度学习实体识别模型从所述诉求信息中提取出有效诉求，并对所述有效诉求进行特征提取，以提取得到有效诉求特征；
    将所述有效诉求特征输入至多个不同的预测模型分别进行科室预测，得到每个所述预测模型输出的科室预测结果；其中，所述预测模型包括深度学习模型、规则匹配模型、药品匹配模型以及疾病匹配模型；
    对全部所述科室预测结果进行过滤和排序，得到排序结果；
    根据所述排序结果推送目标科室信息。
  16. 根据权利要求15所述的计算机可读存储介质，其中，所述将所述有效诉求特征输入至多个不同的预测模型分别进行科室预测，得到每个所述预测模型输出的科室预测结果，包括：
    将所述有效诉求特征输入已训练完成的所述深度学习模型中，得到所述深度学习模型输出的科室预测结果；
    将所述有效诉求特征输入至所述规则匹配模型中，得到所述规则匹配模型输出的科室预测结果；其中所述规则匹配模型是根据所述有效诉求特征，将匹配出正向规则且没有匹配出逆向规则的科室作为科室预测结果；
    将所述有效诉求特征输入至所述药品匹配模型中，得到所述药品匹配模型输出的科室预测结果；其中所述药品匹配模型是根据所述有效诉求特征获取对应的药品特征，将匹配与所述药品特征对应的科室作为科室预测结果；
    将所述有效诉求特征输入至所述疾病匹配模型中，得到所述疾病匹配模型输出的科室预测结果；其中所述疾病匹配模型是根据所述有效诉求特征获取对应的疾病特征，将匹配与所述疾病特征对应的科室作为科室预测结果。
  17. 根据权利要求16所述的计算机可读存储介质，其中，所述深度学习模型包括已训练完成的贝叶斯网络模型和BERT网络模型；所述将所述有效诉求特征输入已训练完成的所述深度学习模型中，得到所述深度学习模型输出的科室预测结果，包括：
    将所述有效诉求特征分别输入至所述贝叶斯网络模型和所述BERT网络模型；
    当所述贝叶斯网络模型的输出结果与所述BERT网络模型的输出结果相同时，将相同的输出结果作为科室预测结果；
    当所述贝叶斯网络模型的输出结果与所述BERT网络模型的输出结果不相同时，将所述贝叶斯网络模型的输出结果和所述BERT网络模型的输出结果分别进行置信度标准化处理，得到处理结果；
    将所述处理结果进行排序，将排序最前的输出结果作为科室预测结果。
  18. 根据权利要求16所述的计算机可读存储介质，其中，所述对全部所述科室预测结果进行过滤和排序，得到排序结果，包括：
    根据所述患者的性别、年龄、偏好以及历史问诊信息对全部所述科室预测结果进行过滤，得到过滤后的所述科室预测结果；
    依次按照所述规则匹配模型、所述疾病匹配模型、所述药品匹配模型以及所述深度学习模型输出的科室预测结果的顺序对所述过滤后的所述科室预测结果进行排序。
  19. 根据权利要求15所述的计算机可读存储介质，其中，所述科室分诊方法还包括：
    当通过所述深度学习实体识别模型从所述诉求信息中没有提取到有效诉求，推送所述诉求信息的补全请求；
    当接收到补全后的诉求信息，通过所述深度学习实体识别模型从所述补全后的诉求信息中提取出有效诉求。
  20. 根据权利要求16所述的计算机可读存储介质，其中，在所述根据所述排序结果推送目标科室信息之后，所述科室分诊方法还包括：
    获取患者的转诊信息；
    通过所述转诊信息对已训练完成的所述深度学习模型进行迭代更新，得到更新后的所述深度学习模型。
PCT/CN2022/087997 2021-08-30 2022-04-20 科室分诊方法、系统、设备以及存储介质 WO2023029514A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111004639.3A CN113707285B (zh) 2021-08-30 2021-08-30 科室分诊方法、系统、设备以及存储介质
CN202111004639.3 2021-08-30

Publications (1)

Publication Number Publication Date
WO2023029514A1 true WO2023029514A1 (zh) 2023-03-09

Family

ID=78656842

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/087997 WO2023029514A1 (zh) 2021-08-30 2022-04-20 科室分诊方法、系统、设备以及存储介质

Country Status (2)

Country Link
CN (1) CN113707285B (zh)
WO (1) WO2023029514A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113707285B (zh) * 2021-08-30 2024-07-26 康键信息技术（深圳）有限公司 科室分诊方法、系统、设备以及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180315182A1 (en) * 2017-04-28 2018-11-01 Siemens Healthcare Gmbh Rapid assessment and outcome analysis for medical patients
CN108766545A (zh) * 2018-05-31 2018-11-06 平安医疗科技有限公司 在线问诊科室分配方法、装置、计算机设备和存储介质
CN108922608A (zh) * 2018-06-13 2018-11-30 平安医疗科技有限公司 智能导诊方法、装置、计算机设备和存储介质
CN110675944A (zh) * 2019-09-20 2020-01-10 京东方科技集团股份有限公司 分诊方法及装置、计算机设备及介质
CN113707285A (zh) * 2021-08-30 2021-11-26 康键信息技术（深圳）有限公司 科室分诊方法、系统、设备以及存储介质

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108986907A (zh) * 2018-07-24 2018-12-11 郑州大学第附属医院 一种基于knn算法的远程医疗自动分诊方法
CN110047584A (zh) * 2019-04-23 2019-07-23 清华大学 基于深度学习的医院分诊方法、系统、装置及介质
CN111785368A (zh) * 2020-06-30 2020-10-16 平安科技（深圳）有限公司 基于医疗知识图谱的分诊方法、装置、设备及存储介质
CN112037909B (zh) * 2020-08-31 2023-08-01 康键信息技术（深圳）有限公司 诊断信息复核系统
CN112015917A (zh) * 2020-09-07 2020-12-01 平安科技（深圳）有限公司 基于知识图谱的数据处理方法、装置及计算机设备
CN112530576A (zh) * 2020-11-30 2021-03-19 百度健康(北京)科技有限公司 一种线上医患匹配方法、装置、电子设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180315182A1 (en) * 2017-04-28 2018-11-01 Siemens Healthcare Gmbh Rapid assessment and outcome analysis for medical patients
CN108766545A (zh) * 2018-05-31 2018-11-06 平安医疗科技有限公司 在线问诊科室分配方法、装置、计算机设备和存储介质
CN108922608A (zh) * 2018-06-13 2018-11-30 平安医疗科技有限公司 智能导诊方法、装置、计算机设备和存储介质
CN110675944A (zh) * 2019-09-20 2020-01-10 京东方科技集团股份有限公司 分诊方法及装置、计算机设备及介质
CN113707285A (zh) * 2021-08-30 2021-11-26 康键信息技术（深圳）有限公司 科室分诊方法、系统、设备以及存储介质

Also Published As

Publication number Publication date
CN113707285B (zh) 2024-07-26
CN113707285A (zh) 2021-11-26

Similar Documents

Publication Publication Date Title
Pezoulas et al. Medical data quality assessment: On the development of an automated framework for medical data curation
Wu et al. Comparison of chest radiograph interpretations by artificial intelligence algorithm vs radiology residents
CN106682397B (zh) 一种基于知识的电子病历质控方法
US20220358410A1 (en) Methods and systems for medical record searching with transmittable machine learning
CN111599488A (zh) 一种智能问诊实现方法、系统和存储介质
CN108766561B (zh) 病症信息处理方法、装置、计算机设备和存储介质
KR102424085B1 (ko) 기계-보조 대화 시스템 및 의학적 상태 문의 장치 및 방법
CN106295182A (zh) 一种基于患者生物信息的身份识别方法
CN113488157B (zh) 智能导诊处理方法、装置、电子设备及存储介质
US20190266243A1 (en) Automated mapping of service codes in healthcare systems
CN113284572B (zh) 多模态异构的医学数据处理方法及相关装置
WO2022012687A1 (zh) 医疗数据处理方法与系统
WO2023124837A1 (zh) 问诊处理方法、装置、设备及存储介质
CN113724858A (zh) 基于人工智能的疾病检查项目推荐设备、方法及装置
CN111292814A (zh) 一种医疗数据标准化的方法及装置
CN112037904A (zh) 在线诊疗数据处理方法、装置、计算机设备和存储介质
WO2021051496A1 (zh) 诊断结果识别、模型训练的方法、计算机设备及存储介质
WO2023029514A1 (zh) 科室分诊方法、系统、设备以及存储介质
CN113657086B (zh) 词语处理方法、装置、设备及存储介质
CN112802598A (zh) 基于语音诊疗数据的实时辅助诊疗方法及系统
CN113077894A (zh) 基于图卷积神经网络的皮肤诊断系统、方法、设备及介质
Ding et al. Large Language Multimodal Models for 5-Year Chronic Disease Cohort Prediction Using EHR Data
CN118013001A (zh) 基于知识库和大语言模型的互动式知识交互系统
Rao et al. AI-enabled clinical decision support system
CN114388123A (zh) 智能辅诊方法、装置、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22862660

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 13/06/2024)